Professional Documents
Culture Documents
Abstract
Spacer fabrics are triple-layer structures consisting of two separate outer and one interfacing monofilament middle layers. The
middle interfacing layer connects the two outer layers together. Incorporation of the middle layer imparts unique transfer
properties to the fabric. In this study, models capable of predicting heat and moisture transfer of warp knitted spacer fabrics
based on artificial neutral network technique were developed. The best predictive model was found by manipulation of
network topology, epoch and activation functions. These models predicted transfer properties of the spacer fabrics using
parameters such as fabric mass, fiber composition, porosity and thickness. Models were trained using thirty commercially
available samples of warp knitted spacer fabrics. The prediction power of obtained models was evaluated with five
commercially available samples out of training data. The error percentage of prediction was less than 10% for both obtained
models which confirmed the validity of the developed artificial neural network models as a reliable tool in prediction the heat
and moisture transfer of warp knitted spacer fabrics.
Keywords
Warp Knitted Spacer Fabrics; Artificial Neural Network; Heat and Moisture Transfer
Introduction
Research and development in the field of technical textiles in recent years has resulted in steady expansion of this
class of textile structures for use in numerous varied applications. Although existing fabric formation technologies
or their modifications can be employed in production of technical textiles, warp knitting is by far the most
preferred manufacturing technology for production of technical textiles. Generally warp knitted spacer fabrics are
comprised of two inter-connected outer layers. The inter-connection of these outer layers is achieved by a
monofilament interfacing middle layer [1].
Warp knitted spacer fabrics not only have excellent transversal compressibility, but also have exceptional heat and
moisture transfer together with acceptable air permeability. In applications such as automotive, medical, apparels,
technical and industrial where comfort is the main selection criterion, these properties are of paramount
importance.
There are many studies about the mechanical properties, simulation and numerical modeling of spacer fabrics. An
experimentally validated geometrical model capable of evaluating porosity and capillarity of spacer fabrics used as
absorbent fabric in medical applications was developed by Davies and Williams [2,3]. Results of Borhani et al.[4]
investigation led to development of a mathematical model that described transfer mechanism of water vapor
produced by sweating from the skin to the outer surface of spacer fabric. Joanne and Pui [5] studied various
characteristics of spacer fabrics including low-stress mechanical properties, air permeability and thermal
conductivity. It was reported that air permeability and thermal conductivity of spacer fabric were closely related to
the fabric density.
Ertekin and Marmarali [6] investigated thermal comfort properties of spacer fabrics produced on double jersey
machine using three different dial setting and two different spacer yarns. This research showed that fabric weight,
thermal conductivity, thermal resistivity, air permeability and relative water vapor permeability properties are
significantly affected by dial setting and the type of spacer yarn. Compression behavior of warp-knitted spacer
Textiles and Light Industrial Science and Technology, Vol. 4, No. 1June 2015
2304-9421/15/01 017-10 2015 DEStech Publications, Inc.
doi: 10.12783/tlist.2015.0401.02
17
18
fabrics was studied by Psilla et.al [7], Sheikhzadeh et.al [8], Liu et.al [9], and Mokhtari et.al [10]. Renkens et.al [11]
worked on geometry modelling of these fabrics. Three-dimensional simulation of warp knitted spacer fabrics was
also proposed by Miaoa et.al [12].
Artificial neural network technique or ANN is a data analysis method based on intelligent technology. ANN
modeling methods are very popular in prediction the properties of textiles based on specific properties and also
raw materials. These methods generally are very accurate and also simple in compare with analytical models.
Moreover, in analytical modeling some assumptions should be considered which decrease the accuracy of
proposed models.
Extensive use of ANN has been made successfully in various textile disciplines such as yarn and fabric
manufacture or determination of fabric properties [13-24]. Works of Beltran et al on the pilling tendency of wool
knits [13] or that of Tokarska on prediction of permeability of woven fabrics [14] are two examples of application of
ANN technique in textiles. Other application of ANN modeling in the field of textiles is concerned with
development of predictive model based on ANN methodology for dyeing of polyester fibers and its comparison
with statistical regression and fuzzy regression [15]. ANN technique has also been used for prediction of cotton
yarn hairiness [16] or tensile properties of cotton-covered nylon core yarns [17]. Artificial neural network model
was proposed to predict the pilling performance of weft knitted fabrics produced from wool/acrylic blended yarns
[18]. Predicting model of tensile properties and breaking elongation of ring spun yarns was developed by Ramesh
et.al [21] and Majumdar et.al [22] respectively. Contrary to variety application of ANN technique in textiles, the
authors did not find any study on prediction of warp spacer knitted fabrics properties by using artificial neural
network modeling technique.
Therefore, in the present work, it is attempted to predict heat and moisture transfer capacity of warp knitted spacer
fabrics by development of ANN models based on mass, thickness, porosity and fiber kind of warp knitted spacer
fabrics.
Material and Methods
Mass and Thickness
Mass and thickness of warp knitted spacer fabrics was measured according to ASTM D3776 [25] and ASTM D1777
[26] standard test methods, respectively.
Porosity
The fabric porosity () is defined as the ratio of the fabric total void volume to fabric total volume. Equation 1
denotes the dependence of fabric porosity on fabric density (fabric) and the fiber density (fiber).
= 1
fabric
fiber
(1)
Modeling the Moisture and Heat Transfer of Warp Knitted Spacer Fabrics Using Artificial Neural Network Algorithm
19
90%RH) and the ambient environment controlled at 25C and 65% RH[27].
(a)
(b)
FIGURE 2. TYPICAL VARIATION OF HEAT AND MOISTURE TRANSFER THROUGH FABRIC (A). HEAT TRANSFER VS. TIME,
(B). MOISTURE TRANSFER VS. TIME
In Figure2 the variation of humidity and temperature against time are shown respectively. During the initial stage
of test, temperature and humidity in the inner and outer surfaces of the fabric are similar to those of environment
condition. Then, evaporation of sweat from the skin simulator causes rapid increase in temperature and humidity
in the microclimate and rapid increase in the temperature and humidity in the inner surface of the fabric occurs.
This results in creation of a thermal gradient between the two faces of sample. Gradual reduction in the rate of heat
and humidity transfer results in creation of state of balance where evaporation rate from the skin simulator equals
the diffusion rate from the fabric. Ultimately temperature and humidity variation reach a static state. By
considering the heat and moisture transfer curve of samples (As shown in Figure 2), moisture transfer percentage
(Equation 2) and heat transfer percentage (Equation 3) can be obtained.
Moisture Transfer Percentage
=
R 300 R i
100
Ri
(2)
Where R300 is defined as humidity transferred through the fabric after 300 seconds and Ri is initial humidity of
sample.
T300 Ti
100
Ti
(3)
20
Where T300 is heat transferred through the fabric after 300 seconds and Ti is initial temperature of sample.
Arttificial Neural Network Modeling
Artificial Neural Network Parameters
In this study, fiber composition, fabric thickness, weight and prosity of warp knitted spacer fabrics were selected as
input parameters of Artificial Neural Network(ANN) models. Fiber type critically affects fabric heat and moisture
properties. Therfore commercial fabrics made of polyester and nylon multifilament yarns were selected for training
and testing process of ANN proposed models. Specifications of selected warp knitted spacer fabrics are shown in
Table 1. According to Equation 4 [28], Fabric Thickness can affect moisture transfer capacity of fabric.
P=
FL
C
(4)
Where P is permeability coefficient, F is the moisture flux, L is the thickness, and is concentration gradient imposed
across the sample. Equation 4 shows that permeability coefficient is directly proportional to thickness.
TABLE 1. SAMPLE SPECIFICATIONS
Input parameters
Thickness (mm)
Mass (g/m )
1 6.8
131.8 -652
Measured Properties
Porosity (%)
88-97.13
Fiber type
Polyester &
polyamid
Heat transfer(%)
28.43 94.76
Moisture transfer(%)
70.9 98.17
The target parameters of models were heat and moisture transfer of the warp knitted spacer fabrics. Traninig of
model was performed with 30 samples and five samples was used as testing data. In Table2 the specification of
testing set is presented. Therefore two separate model for moisture and heat transfer prediction based on four
input parameter was proposed. Each model have four input unit (fiber composition, fabric thickness, weight and
prosity of warp knitted spacer fabrics) and one output neuron (Moisture transfer percentage and Heat transfer
percentage).
Training process of ANN models was carried out using backpropagation learning algorithm technique based on
gradient descent and momentum rate. The learning rate and momentum values were set at 0.05 and 0.9
respectively.
Overfitting of artificial neural network models can be prevented by regularization or early stopping methods. The
selection of either of these methods is dependent on the population of data. In this work due to availability of small
number of data, overfitting of the model was prevented using regularization method. This method is based on
modification of performance function. Equation 5 shows MSEREG function which is the modified function in
Matlab software.
E( w=
) msereg
= mse + (1 )msw
(5)
Where:
msw =
1
w ij2
n i,j
(6)
= Performance ratio which determines the effectivness of MSW(Mean Saquare of connection weights) and
Modeling the Moisture and Heat Transfer of Warp Knitted Spacer Fabrics Using Artificial Neural Network Algorithm
21
Normalizition of input and output data was carried out in a manner that standard deviation of unity and average
value of zero were obtained. Elimination of the effect of different units of input and output parameters is critically
dependent on this preprocessing operation. Figure 3 shows the most common type of transfer fuctions.
Nonlinearity is the main feature of sigmoid and tangent hyperbolic functions. The tangent hyperbolic (Tansig) and
sigmoid (Logsig) function compress the output data between [-1,1] and [0,1] respectively. According to literatures,
applying sigmoid and tangent hyperbolic function in hidden layer(s) and linear function in output layer could
improve the performance of artificial neural network model [30].
Chattopadhyay reported that the artificial neural network model with one hidden layer and sigmoid transfer
function in hidden layer and linear transfer function in output layer could predict each functions [31]. Parallel to
obtain optimum topology of artificial neural network model, the best transfer function,and number of epoch were
obtained by trail and error. Mean square error(MSE) of predicting testing data was considered as optimization
criterian for judjment.
(a)
(b)
(c)
In this study two separate artifiicial neural network model for prediction the moisture and heat transfer capacities
of warp knitted spacer fabrics were developed. Fabric thickness, prosity, mass, and fiber composition were selected
as inputs to the modles. Each model consists of four input parameters and one outpout unit. Since fiber
composition was not qunatitative parameter it was coded as 0 and 1 for nylon and polyester fiber respetively.
Additionally architecture parameters, number of epoch, transfer functions for hidden layer(s) and output layer,
number of hidden layer(s), and number of hidden neurons was optimized.
TABLE 2. SPECIFICATIONS OF TESTING SET SAMPLES
Sample code
Material
1
2
3
4
5
Polyester
Polyester
Polyester
Polyester
Polyester
Input parameters
Thickness
(mm)
6.8
4.8
3.4
3.9
3.7
Mass
(g/m2)
476.62
304.17
291.77
371.16
384.63
Porosity
(%)
94.92
95.41
93.78
93.10
94.47
Target parameters
Moisture
Heat transfer(%)
transfer(%)
31.63
97.97
40.34
98.25
36.68
94.68
31.63
93.6
31.18
95.3
22
hyperbolic and linear transfer function for hidden neurons and output neuron were selected respectively. The
number of epoch was changed from 500 to 1500 at 100 step. Results showed that best prediction performance of
artificial neural network was obtained after 1000 epoch.
TABLE 3. BEST TRANSFER FUNCTION SELECTED FOR NEURONS OF HIDDEN AND OUTPUT LAYERS
Topology
Epoch
4-7-1
4-7-1
4-7-1
4-7-1
4-5-5-1
4-5-5-1
4-5-5-1
4-5-5-1
1000
1000
1000
1000
1000
1000
1000
1000
1st Hidden
Layer
Tansig
Logsig
Tansig
Logsig
Logsig
Tansig
Logsig
Tansig
Output Layer
---------
Linear
Logsig
Tansig
Logsig
Logsig
Tansig
Linear
Linear
Logsig
Tansig
Logsig
Tansig
MSE
Train
20.94
73.99
36.60
49.39
73.83
44.62
60.53
45.29
MSE
Test
46.78
96.11
50.20
66.82
97.09
62.54
54.05
31.06
Sample
Transfer Function
1
2
3
4
5
6
7
8
9
10
11
12
Tansig/ Linear
Tansig/ Linear
Tansig/ Linear
Tansig/ Linear
Tansig/ Linear
Tansig/ Linear
Tansig / Tansig / Linear
Tansig / Tansig / Linear
Tansig / Tansig / Linear
Tansig / Tansig / Linear
Tansig / Tansig / Linear
Tansig / Tansig / Linear
No. of neurons in
2nd Hidden Layer
------------4
5
6
7
8
9
MSE
Train
47.10
29.41
28.05
14.35
23.37
23.69
17.00
29.47
42.47
17.12
14.05
13.75
MSE
Test
49.69
96.63
63.22
23.69
42.65
46.48
47.59
60.72
44.84
25.27
43.18
48.88
Modeling the Moisture and Heat Transfer of Warp Knitted Spacer Fabrics Using Artificial Neural Network Algorithm
23
4-7-1 structure.
2)Number of Epoch
In order to obtain the optimum number of epochs, this parameter was changed between 500 to 1500 at 100 step.
ANN model comprised of two hidden layers with five neurons in each of them. According to previous results,
the transfer functions of hidden layers and output neuron was sigmoid and linear respectively. Results showed
that the best performance of ANN model was obtained at 700 epoch of training.
3)Structure of Artificial Neural Network Model
To find the best structure of ANN model twelve different structures having various numbers of hidden layers
and number of hidden neurons were chosen (As shown in Table 6). Comparison of the results showed that
structure with two hidden layers and eight neurons in first and second hidden layer provided the lowest
prediction error on testing data. In predicting the heat transfer of testing data, the mean square error(MSE) was
6.88.
Results and Discussion
Table 7 shows both experimental and predicted values of moisture transfer of testing data accompained with
prediction errors. As can be seen the maximum and minimum of prediction error are 6.32% and 4.02 % respectively.
The average of prediction error is 5.01%. These results confirms the capability of artificial neural network
methodology in prediction of moisture transfer of warp knitted spacer fabrics. The performance of optimized
artificial neural network model in prediction of heat transfer capacity of warp knitted spacer fabrics is shown in
Table 8. Results shows that average of prediction error is 6.36% and the maximum and minimum of prediction
error are 9.54 % and 3.60% respectively. Again the predictive power of artificial neural network model in
prediction of moisture capacity of warp knitted fabrics is confirmed.
TABLE 5. BEST TRANSFER FUNCTION SELECTED FOR HIDDEN AND OUTPUT NEURONS
Topology
4-7-1
Tansig
---
4-7-1
Logsig
---
4-7-1
Tansig
---
4-7-1
Logsig
---
MSE
Train
MSE
Test
Linear
6.69
31.52
Linear
13.57
20.10
Tansig
12.55
46.76
Logsig
38.21
95.69
57.52
Output Layer
4-5-5-1
Tansig
Tansig
Linear
5.20
4-5-5-1
Logsig
Logsig
Linear
7.41
26.37
4-5-5-1
Tansig
Tansig
Tansig
11.17
108.54
4-5-5-1
Tansig
Logsig
Logsig
31.78
87.69
TABLE 6. PERFORMANCE OF DIFFERENT TOPOLOGIES AFTER 700 EPOCH ON TESTING AND TRAINING DATA SET
Sample
Transfer Function
No. of neurons in
1st Hidden Layer
---
No. of neurons in
2nd Hidden Layer
MSE
Train
MSE
Test
Logsig / Linear
21.43
21.43
Logsig / Linear
---
16.06
25.78
Logsig / Linear
---
18.23
21.07
Logsig / Linear
---
10
25.47
62.36
Logsig / Linear
---
12
22.72
72.00
Logsig / Linear
---
14
22.44
106.06
35.61
114.08
21.64
65.27
21.77
47.35
10
18.05
29.37
11
5.83
6.88
12
18.60
68.94
24
Sample
Target Value
Predicted Value
Error
Error(%)
97.97
94.03
-3.94
4.02
98.25
93.23
-5.02
5.11
94.68
89.40
-5.28
5.57
93.63
87.68
-5.92
6.32
95.34
91.45
-3.85
4.04
Sample
Target Value
Predicted Value
Error
Error(%)
31.63
32.78
1.15
3.60
40.334
43.38
3.04
7.53
36.68
40.18
3.50
9.54
31.63
29.55
-2.08
6.56
31.18
29.76
-1.42
4.57
Conclusion
In this study heat and moisture transfer properties of warp knitted spacer fabrics were predicted using artificial
neural network algorithm. Strucutral parameters of fabrics were used as input of developed models. Results
confirmed the ability of ANN modeling technique as an valuable method in prediction of transfer phnomenon of
warp knitted spacer fabrics. It was found that topolgy with one hidden layer and ten neurons, after 700 epoch,
yielded the best performance for heat transfer predcition. Similarly it was found that topology with two hidden
layers and eight neurons, after 1000 epoch provided the best moisture transfer prediction. The mean square error of
testing data for heat transfer and moisture transfer were found to be 23.69 and 6.88 respectively. Additionally it
was concluded that the maximum predction error for hear and moisture transfer were 6.32% and 9.54%
respectively.
REFERENCES
[1]
Liu, Y., Sun, B., Hu, H., Gu, B., Dynamic Response of 3D Biaxial Spacer Weft- knitted Composite under Transverse
Impact, J. of Reinforced Plastics and Composites, Vol. 25, No. 15, 2006, pp.1629-1641.
[2]
Davies, A. and William, J., The use of spacer fabrics for absorbent medical applications, J. of Fiber Bioengineering and
informatics, Vol. 1, No. 4, 2009, pp.321-330.
[3]
Delkumbureatte, G. and Dias T., porosity and capillary of weft knitted spacer structure, Fiber and polymers, Vol. 10, No. 2,
2009, pp.226-230.
[4]
Borhani, S., Seirafianpour, S., Hosseini, S.A., Sheikhzadeh, M., Mokhtari, M., Computational and Experimental
[5]
Spacer Fabrics, J. of Engineered Fibers and Fabrics, Vol. 5, No. 3, 2010, pp.43-48.
[6]
Joanne, Y. and Sun-Pui, N., Study of three-dimensional spacer fabrics: Physical and mechanical properties, J. of Materials
[7]
[8]
Ertekin, G. and Marmali, A. Heat, Air and Water Vapor Transfer Properties of Circular Knitted Spacer Fabrics, Tekstil ve
Vasiliadis, S., Kallivertaki, A., Psilla, N., Provatidis, CH., Mecit, D., Roye, A., "Numerical modelling of the compressional
behaviour of warp-knitted spacer fabrics", Fibers & Textiles in Eastern Europe, Vol. 17, No. 5, 2009, pp.56-61.
[10] Sheikhzadeh, M., Ghane, M., Eslamian, Z., Pirzadeh, E., A modeling study on the lateral compressive behavior of spacer
fabrics, J. of the Textile Institute, Vol. 101, No. 9, 2010, pp.795-800.
[11] Liu, Y., Hu, H., Long, H., Zhao, L., Impact compressive behavior of warp-knitted spacer fabrics for protective
applications, Textile Research J., Vol. 82, No. 8, 2012, pp.773-788.
Modeling the Moisture and Heat Transfer of Warp Knitted Spacer Fabrics Using Artificial Neural Network Algorithm
25
[12] Mokhtari, F., Shamshirsaz, M., Latifi, M., Maroufi, M., Compressibility behavior of warp knitted spacer fabrics based on
elastic curved bar theory, J. of Engineered Fibers and Fabrics, Vol. 6, No. 4, 2012, pp.23-33.
[13] Renkens, W. and Kyosev, Y., "Geometry modeling of warp knitted fabrics with 3D form." Textile Research J., Vol. 81, No. 4,
2011, pp. 437-443.
[14] Zhang, L.Z., Jiang, G.M., Miao, X.H., Conga, H.L., "Three-dimensional Computer Simulation of Warp Knitted Spacer
Fabric", Fibres & Textiles in Eastern Europe,
[15] Beltran, R., Wang, L., Wang, X., Predicting the Pilling Tendency of Wool Knits,
2, 2006, pp.129-136.
[16] Tokarska, M., Neural Model of the Permeability Features of Woven Fabrics, Textile Research J., Vol. 74, No. 12, 2004,
pp.1045-1048.
[17] Nasiri, M., Shanbeh, M., Tavanai, H.,
Network Modeling Methodologies in Polyester Dying, Proceeding of International Conference on Computational intelligence
for Modeling, Control and Automation, 2005, pp.505-510, Austria.
[18] Babay, A., Cheikhrouhou, M., Vermeulen, B., Rabenasolo, B., Castelain, J.M., Selecting the optimal neural network
architecture for predicting cotton yarn hairiness, J. of the Textile Institute, Vol. 96, No. 3, 2005, pp.185-192.
[19] Ghareaghaji, A.A., Shanbeh, M., Palhang, M.,Analysis of Two Modeling Methodologies for Predicting the Tensile
Properties of Cotton-Covered Nylon Core Yarns, Textile Research J., Vol. 77, No. 8, 2007, pp.565-571.
[20] Hasani, H. and Shanbeh, M., "Application of multiple linear regression and artificial neural network algorithms to predict
the total hand value of summer knitted T-shirts", Indian J. of Fibre & Textile Research, Vol. 35, No. 3, 2010, pp.222-227.
[21] Rejali, M., Hasani, H., Ajeli, S., Shanbeh, M.,Optimization and prediction of the pilling performance of weft knitted
fabrics produced from wool/acrylic blended yarns, Indian J. of Fibre & Textile Research, Vol. 39, No. 1, 2014, pp.83-88.
[22] Vadood, M., Safar Johari, M., Rahai, A., "Developing a hybrid artificial neural network-genetic algorithm model to predict
resilient modulus of polypropylene/polyester fiber-reinforced asphalt concrete", J. the Textile Institute, online printed, 2014,
pp.1-12.
[23] Ramesh, M.C., Rajamanickam, R., Jayaraman, S., The prediction of yarn tensile properties by using artificial neural
networks", J. the Textile Institute, Vol. 86, No. 3, 1995, pp. 459-469.
[24] Majumdar, P.K. and Majumdar, A., Predicting the breaking elongation of ring spun cotton yarns using mathematical,
statistical, and artificial neural network models", Textile Research J., Vol. 74, No. 7, 2004, pp.652-655.
[25] Luo, C. and Adams, D.L., "Yarn strength prediction using neural networks part I: fiber properties and yarn strength
relationship", Textile Research J., Vol. 65, No. 9, 1995, pp.495-500.
[26] reyen, M. and Pelin, G., Comparison of artificial neural network and linear regression models for prediction of ring
spun yarn properties. I. Prediction of yarn tensile properties", Fibers and Polymers, Vol. 9, No. 1, 2008, pp.87-91.
[27] ASTMD3776, 1997. Standard Test Method for Thickness Textile Material, American Society for Testing and Materials.
[28]
ASTMD1777, 1997. Standard Test Method for Mass per Unit Area of Fabric, American Society for Testing and Materials.
[29] Ziaee, M., Borhani, S., Shanbeh, M., Evaluation of physical and mechanical properties of cotton covered polypropylenecore yarns and fabrics, Industria Textila, Vol. 62, No.1, 2011, pp. 9-13.
[30] Wehner, J.A., Miller, B., Rebenfeld, L., Dynamic of Water Vapor Transmission Through Barriers, Textile Research J.,Vol.
58, No. 10, 1988, pp.581-592.
[31] Demuth, H. and Beale, M. 2001. Neural Network Toolbox/Matlab Software, the Mathworks Inc.
[32] Gurney, K. 1997. An introduction to Neural Network, UCL Press, London.
[33] Chattopadhyay, R., Application of Neural Network in Yarn Manufacture, Indian J. of Fiber and Textile Research,Vol. 31,No.
1, 2006, pp.160-169.
26
Dr. Mohammad Zarrebini obtained his B.Sc in textile technology from University of Leeds. He obtained his M.Sc from Leeds
polytechnic and Ph.D from Leeds university in nonwovens. He is currently head of textitle engineering department at Isfahan
university of technology. He has published miscellaneous papers in fields of various textiles structures and filters.
Dr. Hossein Hasani is presently working as an Associate Professor at Department of Textile Engineering at Isfahan university
of technology, Isfahan (Iran). He did his M.Sc. in textile engineering from Amirkabir university, Tehran (Iran) and Ph.D. from
Stuttgart university, Stuttgart (Germany). He has 10 years of research experience including seven years of teaching spinning
and knitting technologies for graduate students. Dr. Hasani has published 40 research papers in peer reviewed international
and national journals.